How CACI built a secure, scalable AI sandbox for a financial services company to modernise data capabilities

Solution

CACI approached this as a security-first architecture challenge, designing a sandbox that would enable AI experimentation while meeting the company’s strict regulatory and governance requirements.

Secure-by-design, enterprise-grade architecture

• A standard Azure Machine Learning deployment can be created with only a handful of CLI commands. To satisfy the client’s security requirements, we designed a fully private architecture with no public endpoints, adding virtual networking, private DNS, network security controls, private connectivity and secure access mechanisms. The resulting infrastructure was codified in over 1,300 lines of Bicep.

Fully scripted, repeatable infrastructure (Infrastructure-as-Code)

• Every element, from network configuration to compute provisioning, was defined in code and version-controlled. This ensures environments can be reliably recreated, audited and scaled without deviation or manual risk.

Tightly governed, private environment

• The sandbox was built with private endpoints and no public network exposure, embedding security from the ground up and ensuring alignment with the client’s stringent data protection requirements.

Integrated, flexible tooling for users

• The environment supports Azure Machine Learning Studio alongside VS Code, Jupyter and RStudio, enabling both data scientists and wider teams to work effectively within a single, governed ecosystem.

Enablement through clear guidance and documentation

• CACI delivered practical Quickstart guides and documentation, equipping the company’s teams to confidently use, extend and replicate the environment, turning a one-off build into a repeatable organisational capability.

Results

The outcome was a working sandbox with a trusted foundation for AI innovation within a highly regulated context.

By CACI introducing a fully repeatable, code-driven approach, the company can now create new environments quickly and consistently, without reintroducing risk. This has significantly reduced the operational overhead associated with setup while improving alignment across development, test and future production environments.

Just as importantly, the solution brought clarity and control. Version-controlled infrastructure and standardised configurations provide full transparency over how environments are built and managed, supporting auditability and governance processes.

From a user perspective, the improved tooling and clearer workflows have made it easier to work with models, accelerating experimentation and enabling broader engagement beyond purely technical teams.

The impact has been both immediate and strategic. The client not only successfully established a secure sandbox but also gained a reusable approach that can support future AI use cases with confidence.

This confidence was reflected in continued engagement, with the company extending CACI’s support and exploring how the sandbox can underpin additional, production-facing initiatives.